8 CIRCULAR 3 3 3, U.S. DEPARTMENT OF AGRICULTURE 



another set of plots established, so far as they could judge, under the 

 same conditions. Too often they have, in fact, made sweeping de- 

 ductions regarding the effect of a single treatment in the absence of 

 dependable evidence that the treatment had any significant effect. 



Defects are inevitable in a comparison of treatments based on a 

 single pair of plots. Uncontrolled factors inherent in climate, in 

 soil conditions, in slope or other factors of site, in biological con- 

 ditions, or in stand history frequently have a greater part in the 

 apparent response to treatment than the controlled factors, or at 

 least create differences that partially or entirely mask the differences 

 due to the factors under control. The only remedy is to make enough 

 replications to test the significance of the observed differences. 

 There is safety in numbers. 



Replication serves two purposes: it effects a high degree of pre- 

 cision by diminishing the standard error of the experimental com- 

 parisons, and it provides an estimate of error. Both of these are 

 needed in any experiment. The former can be obtained by care in 

 the conduct of the work, but the latter, essential to any reliable 

 scientific study, can be obtained by no means other than replication. 

 The latter purpose has often been overlooked, partly because so much 

 emphasis was given the former. 



In silvicultural research, satisfactory replications can sometimes 

 be obtained by dividing large plots and treating the data from each 

 division as if they came from a separate plot. One advantage of 

 dividing plots is that it helps to detect unusual conditions existing 

 on one part of a plot and not on others. The desirable size and 

 number of divisions depend upon the uniformity of conditions on 

 the plot as a whole and upon the conditions and purpose of the 

 experiment. For example, a plot of one-tenth of an acre in dense re- 

 production might well be divided into 10 parts, while a plot of 100 

 acres in a mature and open pine stand might not be suited to divi- 

 sion. The possibilities that lie in plot division have not been fully 

 explored in silvicultural research. It must be remembered that 

 division does not eliminate variables due to time. 



Before plots are established the stand should be thoroughly ex- 

 amined for uniformity. Obviously, the more nearly conditions in 

 the pairs of treated and control plots are alike before treatment, 

 the fewer are the replications needed to obtain a result of a given 

 degree of accuracy. The kinds of variation most frequently over- 

 looked include differences in average diameter, density, diameter dis- 

 tribution, stem distribution, species composition, basal area, volume, 

 site, and soil conditions (type, depth, texture, and drainage). Plots 

 established one above another on a slope are unlikely to yield com- 

 parable data, because of differences in soil, site, or drainage entailed 

 by such relative location. Plots on south slopes are usually unsuit- 

 able to be compared with plots on north slopes, because of differences 

 in such factors as humidity, frost, snow accumulation and melt, and 

 length of growing season. 



Response to treatment may be profoundly affected if the treat- 

 ment is given during abnormally dry or abnormally wet years, or 

 at a time when weather conditions are at a peak of cyclic changes. 



